Epigenetic silencing of TGFBI confers resistance to trastuzumab in human breast cancer

SKBr3 (SK) and AU565 (AU) HER2+ breast carcinoma cells were obtained from Eucellbank (University of Barcelona, Spain) [15] and the American Type Culture Collection (ATCC, Rockville, MD, USA). SKBr3 and AU565 cells were routinely grown in McCoy’s (Gibco) and Dulbecco’s modified Eagle’s medium (DMEM; Gibco), respectively, and supplemented with 10% FBS (HyClone Laboratories), 1% l-glutamine (Gibco), 1% sodium pyruvate (Gibco), and 100 U/mL penicillin/streptomycin (HyClone Laboratories). Cell lines were kept at 37 °C and 5% CO₂ atmosphere. Long-term trastuzumab-resistant SK cells (SKTR) and AU565 cells (AUTR) had been previously developed in our laboratory [16, 17]. Resistance was confirmed with cell viability assays. The trastuzumab-resistant SKTR and AUTR cells were maintained in 2 μM of trastuzumab, i.e., a concentration in which parental cells were not viable.

TGFBI promoter methylation levels were retrospectively evaluated in tumor samples from 24 patients diagnosed with HER2+ BC at the Dr. Josep Trueta University Hospital, Girona (Spain) between 2007 and 2015. The patients were selected from the hospital’s pharmacy database. The selection criterion included patients with early or locally advanced HER2+ BC who had received neoadjuvant treatment with trastuzumab and chemotherapy. Twenty patients had no response or partial response and 4 patients had complete response to trastuzumab plus chemotherapy. For all patients, hematoxylin and eosin (H&E)-stained slides from formalin-fixed paraffin-embedded (FFPE) tumor blocks were examined to determine the representative areas of the invasive tumor. Estrogen receptor (ER), progesterone receptor (PR), and HER2 expression had been previously analyzed in the tumors using immunohistochemistry (IHC). For each patient, clinical and histopathological features were obtained: age, stage (TNM classification [18]), histological grade (Bloom-Richardson grading system), menopause status, type of surgery, and relapse.

Epigenetic signatures are characterized by a very dynamic nature, where DNA methylation has often been shown as a reversible mechanism of transcriptional control by inhibition of enzymes such as the DNA methyltransferases [19]. We performed reactivation treatments using the demethylating agent, 5-aza-2′-deoxycytidine (5-aza-dC, Sigma-Aldrich, St Louis, MO) at 3 μM and 5 μM of 5-aza-dC for 72 h. The medium was changed every day to promote DNA demethylation.

Genomic DNA extraction from cell lines or formalin-fixed paraffin-embedded (FFPE) core biopsies (10 μm) and tissue sections (5 μm) using a QIAamp DNA Mini Kit and Deparaffinization Solution with a QIAamp DNA FFPE Tissue Kit (Qiagen, Hilden, Germany), respectively, were carried out following the manufacturer’s instructions. For the RT-PCR experiments, cells were washed with PBS and then suspended in 1 mL of Qiazol (Qiagen Hilden, Germany) was added. Total RNA was isolated using a GeneJET RNA Purification Kit (Thermo Fisher Scientific) following the instructions provided by the manufacturer. All DNA and RNA samples were quantified using a NanoDrop 2000 Spectrophotometer (Thermo Fisher Scientific).

DNA was bisulfite-modified using an EZ DNA Methylation-Gold Kit (Zymo Research) in accordance with the manufacturer’s recommendations.

Genome-wide DNA methylation analysis was performed using an Illumina 450K DNA methylation microarray (Infinium HumanMethylation450 BeadChip) as previously described [20]. This technique is an epigenomic approach that allows analyzing the methylation profile of the human genome. Bisulfite-converted DNA from SK and SKTR models were used to hybridized on an Illumina Infinium HumanMethylation450 (450K) BeadChip array, following the Illumina Infinium HD Methylation protocol. Sex chromosomes were excluded from the analysis because they usually represent a high source of variation in DNA methylation levels [21]. Data was deposited into the NCBI Gene Omnibus, accession number: GSE123754.

DNA methylation is an epigenetic mechanism that usually affects CpG islands and promoters regulating the expression of the genes. For this reason, a transcriptomic analysis by RNA-Seq was performed. Library construction was performed using an Illumina TruSeq Stranded mRNA Sample Preparation Kit (CAT. No. RS-122-2101, RS-122-2102). Following the transfer of the flow cell to an Illumina HiSeq instrument, a 101-cycle paired-end sequence run was performed using the HiSeq SBS Kit v4 sequencing reagents (FC-401-4002). A detailed description of the analysis is provided in the Additional file 1. Sequencing data have been posted in the Gene Expression Omnibus (GEO) database under accession number GSE114575.

To validate the results obtained from the arrays, bisulfite pyrosequencing analyses were performed. The primers for the PCR amplification and sequencing were designed with the PyroMark assay design software version 2.0.01.15. DNA-bs was amplified by PCR under standard conditions with biotinylated primers (see Additional file 1: Table S1) to convert the PCR product into single-stranded DNA templates. We used a Vacuum Prep Tool (Biotage, Sweden) to prepare single-stranded PCR products following the manufacturer’s instructions. The PCR products were observed in 2% agarose gels before pyrosequencing. Reactions were performed in a PyroMark Q96 System version 2.0.6 (Qiagen, Hilden, Germany) using appropriate reagents and protocols. The methylation value was obtained from the average of the CpG dinucleotides included in the sequences analyzed using a Pyro Q-CpG 1.0.9 (Qiagen, Hilden, Germany).

DNA methylation was also validated using the methylation-specific PCR (MSP) using a set of primers designed by the Methyl Primer Express program for each gene analyzed. For each sample, two PCR reactions were produced using the same DNA template: one reaction with methylated primers (M: methylated) and the other with unmethylated primers (U: unmethylated). Commercial methylated human male genomic DNA (CpGenome Universal Methylated DNA, Millipore) was used as a methylated positive control (IVD), and DNA from normal lymphocytes (NL) as a positive control for unmethylated alleles. Primer sequences can be found in Additional file 1: Table S2.

To confirm the differential regulation of TGFBI, CXCL2 and SLC38A1 observed in SK and SKTR models used gold standard methodologies for gene expression such as quantitative real-time PCR. Total RNA was reverse-transcribed into cDNA using a High Capacity cDNA Archive Kit (Applied Biosystems). Gene expression levels of selected genes were assessed using a LightCycler 480 Real-time PCR System (Roche) with a LightCycler 480 SYBR Green I Master (Roche). Primers are depicted in Additional file 1: Table S3. RT-PCR analyses were performed at least four times, and each gene was run in triplicate. GAPDH was used as an endogenous control to enable normalization.

For the TGFBI long-term knockdown, two different ShRNAs were specifically designed against TGFBI mRNA (NM_000358) in two different loci, by considering a 19-base target sequence and inserted in the pLVX-shRNA2 vector (Clontech). For stable overexpression, the cDNA sequence of TGFBI was amplified from SKWT cDNA and cloned into the pLVX-IRES-tdTomato vector (Clontech). This construction (wild type) was then used as a template to create its mutated version following a PCR-based strategy. Mutations were performed in the RGD domain and the NKDIL (amino acids 354–358), YH18 (amino acids 563–580), and EPDIM (amino acids 617–621) motif in the second and fourth FAS-1 domains [22]. Lentiviral production was performed using the protocol described in Additional file 1. After lentiviral transduction, ZsGreen1-positive or tdTomato-positive cells were sorted by flow cytometry.

Cell viability was determined using a 3-(4, 5-dimethyl-2-thiazolyl)-2, 5-diphenyl-2 h-tetrazolium bromide (MTT) assay. Briefly, cells were plated in their growth medium in 96-well plates at a cell density of 1.5 × 10³ cells per well. After 24 h, the growth medium was removed, and, 100 μL of fresh medium containing the corresponding concentration of trastuzumab was added to each well for 5 days. Cell viability was measured using the standard colorimetric MTT assay as previously described [16]. Using the multi-well plate reader Benchmark Plus (Bio-Rad), absorbance was determined to be 570 nm. Data presented are from three separate wells per assay, and the assay was performed at least three times.

We previously demonstrated that different activation patterns of some HER receptors and their downstream signaling proteins are associated with the molecular mechanisms of acquired trastuzumab resistance (SK and SKTR) [16]. For this reason, a panel of the different proteins were analyzed in our models using Western blot analysis. The parental (SK and AU) and resistant (SKTR and AUTR) models were synchronized by starvation in serum-deprived medium (0.5% FBS) for 24 h. Cells were lysed in ice-cold lysis buffer (Cell Signaling Technology, Inc.) with 100 μg/mL phenylmethylsulfonylfluoride (PMSF). Protein concentration was determined with Lowry (DC Protein Assay, Bio-Rad). Equal amounts of protein were heated in LDS Sample Buffer with Sample Reducing Agent (Invitrogen) for 10 min at 70 °C, separated on SDS-PAGE and transferred to nitrocellulose membranes. Protein was detected using primary antibodies (Additional file 1: Table S5). Secondary antibodies, α-tubulin and β-actin, conjugated to horseradish peroxidase were used. The immune complexes were detected using a chemiluminescent HRP substrate [SuperSignal™ West Femto Maximum Sensitivity Substrate (Thermo Scientific™ Inc.) or Immobilon Western HRP Substrate (Millipore Sigma)] and in a Bio-Rad ChemiDoc™ MP Imaging System. Western blot analyses were repeated at least three times and representative results are shown.

Data were analyzed with Student’s t test when two groups were being compared or with one-way analysis of variance (ANOVA) followed by Tukey’s honest significant difference (HSD) post hoc test or Tamhane’s T2 post hoc test for multiple comparisons. The non-parametric Kruskal-Wallis test was used when data did not follow normal distribution. When two groups were being compared, the data were analyzed with the Mann-Whitney U tests for non-normally independent variables; otherwise, the Kruskal-Wallis test was used for more than two groups. In the patient cohort, we analyzed the TGFBI promoter methylation status and its association with the clinical-histopathological characteristics. Patient data were summarized as median (first quartile-third quartile) for continuous variables, and frequencies and percentages for categorical variables. The potential association between clinical-histopathological characteristics and levels of TGFBI methylation (low ≤ 20%, high ≥ 20%) or differences in TGFBI methylation before and after treatment were analyzed using the Mann-Whitney U and Kruskal-Wallis tests for continuous variables and the chi-square or Fisher exact tests, as appropriate, for categorical variables. The correlation between variables was observed using Spearman’s rho coefficient. Characteristic (ROC) curves were used to assess the predictive capacity of the TGFBI marker. ROC curve analysis was performed using MedCalc software Version 18.11.3. The Hosmer-Lemeshow test and calibration plot was assessed with MedCalc and STATA software. Levels of significance were set at p < 0.05 and are represented by asterisks, as follows: p < 0.05 (denoted as *), p < 0.01 (denoted as **), and p < 0.001 (denoted as ***). The statistical analysis was performed using the IBM SPSS software (Version 21.0; SPSS Inc.)

The Infinium HumanMethylation 450 BeadChip (450k array) is a validated tool for carrying out epigenomic projects because it allows the methylation status of approximately 450,000 CpGs located throughout the human genome to be detected [20]. Our group recently employed this approach in previous studies aimed at identifying genes regulated by DNA methylation for clinical applications such as biomarkers [23, 24]. Therefore, we took advantage of the 450K array methodology to characterize the DNA methylation profile associated with trastuzumab resistance in BC, and compared a trastuzumab-sensitive (SK) and trastuzumab-resistant (SKTR) BC models (Fig. 1a). The global analysis of the DNA methylation (β values) corresponding to the CpG sites with p value < 0.01 (469,927 CpGs) showed differences in the scatter plot (Fig. 1b) between SK and SKTR models (r² = 0.93) and revealed 27,314 differentially methylated CpGs (Δβ ≥ 0.20) between both models. In particular, the number of CpGs that gained (red triangle in scatter plot) and lost (green triangle in scatter plot) a methylation level ≥ 0.20 in SKTR with respect to SK was 14,845 and 12,469, respectively.

Therefore, we focused our study on analyzing the methylation levels of the CpG sites located at the regulatory regions corresponding to 89,168 CpGs and 13,508 genes. The supervised hierarchical clustering of the most variable CpGs from promoters and islands (Δβ ≥ 0.20) showed a methylation pattern that clearly discriminated between the SK and SKTR models (Fig. 1c). Next, we used more stringent criteria to determine the most differentially methylated CpGs in the promoters and islands, considering in SKTR the CpGs with a methylation level (β) in SK < 0.20 and in SKTR > 0.60 as hypermethylated, and in SKTR the CpGs with a methylation level (β) in SKTR < 0.20 and in SK > 0.60 as hypomethylated. This analysis revealed 184 differentially methylated CpGs corresponding to 152 genes which, according to a Gene Ontology analysis (GO), were significantly associated (FDR < 0.05) with several biological processes related to cancer, such as cell adhesion pathways (GO:0007155), regulation of transcription (GO:0006355), development (GO:0007275), and control of apoptosis (GO:0043065) (Fig. 1d). To verify whether the methylation observed in these genes had an impact on their transcriptional expression or not, we performed a transcriptomic analysis with RNA-Seq comparing the SK and SKTR models and obtained 1995 overexpressed and 3191 downregulated genes in SKTR displaying a fold change ≥ 1.5 over the SK model. Then, we correlated the genes differentially methylated at the promoter and island with those differentially expressed between SK and SKTR (Fig. 1e). We identified 31 hypermethylated and downregulated genes and no gene hypomethylated and overexpressed in SKTR with respect to the SK model (Table 1). From this list of 31 genes, we selected three genes: TGFBI (transforming growth factor beta induced), CXCL2 (C-X-C motif chemokine ligand 2), and SLC38A1 (Solute carrier Family 38 Member 1) for further analysis. TGFBI, CXCL2, and SLC38A1 were selected because of the high number of differentially methylated CpGs (≥ 3 CpG sites) they presented and their previous implications in BC [25‐27].

First, we analyzed the DNA methylation of the selected genes by bisulfite pyrosequencing, which is a technique that allows single genes to be analyzed on a single CpG level [28]. We observed significantly higher methylation levels for the three genes in SKTR relative to the SK (Fig. 2a; p < 0.05 for TGFBI and SLC38A1; p < 0.01 for CXCL2). Like the previous transcriptomic analysis with RNA-Seq, the SKTR model also showed a significant reduction in the transcript expression levels of TGFBI, CXCL2, and SLC38A1 (p < 0.05) when compared to the SK cells analyzed by qRT-PCR (Fig. 2b). Interestingly, TGFBI was the gene that showed the greatest decrease (fold change = 6.7) in transcriptional levels in the SKTR model. Similar results were obtained by methylation-specific polymerase chain reaction (MSP) analysis (a very specific and sensitive method [29]) that showed higher methylation in SKTR with respect to the SK cells for all the genes evaluated (Fig. 2d). These results confirmed that hypermethylation of the three selected genes at the promoter and CpG island in SKTR cells is associated with reduce transcriptional levels.

In this sense, after treating the SKTR model with the demethylating agent 5-aza-2′-deoxycytidine (5-aza-dC) at 3 μM and 5 μM (Fig. 2c), the transcriptional levels of TGFBI, CXCL2, and SLC38A1 were significantly restored (p < 0.05), indicating that methylation is an epigenetic mechanism that has a functional role in the transcriptional control of the three genes in question. Importantly, this epigenetic silencing by DNA methylation observed was also confirmed at the protein level in two (TGFBI and CXCL2) of the three selected genes (Fig. 2e), showing a reduction of the protein levels in SKTR with respect to SK cells that was significantly restored in the hypermethylated SKTR model after the treatment with 5-aza-dC 3 μM and 5 μM. The change observed in the protein levels after the 5-aza-dC treatment was especially drastic in TGFBI gene, where the absence of protein expression induced by the promoter hypermethylation in SKTR was completely recovered after the in vitro demethylation.

Therefore, to verify that the results obtained for these genes were not specific to the SK and SKTR models, we extended their epigenetic analyses to other trastuzumab-sensitive (AU) and trastuzumab-resistant (AUTR) human BC cell models (Fig. 2f and Additional file 2: Figure S1). Likewise, only TGFBI gene showed a high correlation between methylation, expression, and protein levels in the AUTR model. With MSP, we observed that hypermethylation of TGFBI in AUTR (compared to AU) was associated with a significant decrease (p < 0.05) in the transcriptional and protein levels of the gene. Importantly, the transcriptional and protein levels of TGFBI gene were both restored in the hypermethylated AUTR model after the treatment with 5-aza-dC 3 μM and 5 μM. These results confirm that the epigenetic silencing of TGFBI by DNA methylation is associated to trastuzumab resistance in human BC cells.

We have shown that TGFBI promoter hypermethylation is highly associated with the downregulation of its transcript and protein levels in two trastuzumab-resistant models when compared to the corresponding sensitive models. Therefore, we next examined the functional contribution of epigenetic inactivation of the TGFBI gene to trastuzumab resistance. We depleted the endogenous TGFBI gene expression in the SK model by stable transfection with two different short hairpin RNAs (shTGFBI A and shTGFBI B). We also rescued the TGFBI expression in the TGFBI-hypermethylated and the downregulated SKTR models by stable transfection with a plasmid encoding the full-length TGFBI cDNA (TGFBI).

TGFBI is an extracellular matrix (ECM) protein and plays a role in mediating cell adhesion to the ECM, cell proliferation, adhesion, migration, and differentiation through interacting with collagen, fibronectin, laminin, and several integrins [30‐32]. These integrin-binding properties of TGFBI have been related to different integrin-binding motifs located in fasciclin-1 domains, including NKDIL, YH18, and EPDIM as well as an Arg-Gly-Asp (RGD) domain [22]. For this reason, we became interested in determining whether this interaction between TGFBI with the ECM and integrins could be involved in the trastuzumab resistance in our models. With this objective in mind, we also expressed a mutated form of TGFBI with four different altered integrin binding motifs (NKDIL, YH18, EPDIM, and RGD), affecting cellular adhesion through the integrin interactions [22]. The transfection and validation protocols were the same as for the overexpression vector.

Despite its known function in cell adhesion and integrin-mediated signaling, we did not observe morphological changes after depletion, overexpression, or mutagenesis of TGFBI (Fig. 3a). The efficiency of the transfection was assessed by measuring the TGFBI gene expression using quantitative RT-PCR and Western blotting (Fig. 3b). A significant change in its levels was detected both after depletion and overexpression. Upon TGFBI transfection, we analyzed the resistance or sensitivity to trastuzumab in comparison to the parental and empty vector-transfected cells through an MTT assay over 5 days (Fig. 3c; Additional file 3: Figure S2). In the SK model, we observed that TGFBI depletion did not affect the trastuzumab response (Fig. 3c, upper panel). In contrast, overexpression of TGFBI in SKTR led to a significantly higher sensitivity at a trastuzumab concentration of 10 μM compared to the resistant empty vector cell line (Mock: p = 0.020) and the SKTR model (p = 0.004) (Fig. 3c, middle panel). The TGFBI mutated form (TGFBImut) showed the same response to trastuzumab treatment as the mock vector did, suggesting that TGFBI-mediated trastuzumab sensitivity requires the NKDIL, YH, EPDIM, and RGD binding motifs (Fig. 3c, lower panel).

Therefore, we examined HER family protein receptors and their downstream proteins related to PI3K/AKT and MAPK/ERK1/2 pathways for changes after TGFBI depletion, overexpression, and mutation (Fig. 3d). Consistent with the results of the trastuzumab cell viability, TGFBI depletion in SK cells showed no apparent changes in either total protein or activation levels for any of the proteins examined. Unlike TGFBI depletion, overexpression of TGFBI and TGFBImut in the SKTR model produced changes in the activation levels of some of the HER receptors and downstream signaling proteins in comparison with the mock control vector. In particular, overexpression of either wild type TGFBI or its mutated form resulted in a significant increase in the activation levels of HER1 (pHER1), HER2 (pHER2), and AKT (pAKT) compared to the SKTR model (Mock) that contains hypermethylated TGFBI without change in total levels of the respective proteins. Hence, TGFBI overexpression and its mutated form allow the trastuzumab-resistant model to adopt similar activation levels for HER1, HER2, and AKT as the trastuzumab-sensitive cells do.

Given the previous in vitro results obtained, we wanted to determine whether the presence of the TGFBI promoter CpG island hypermethylation-associated to trastuzumab resistance also occurring in BC patients. Therefore, we evaluated TGFBI gene methylation in human tumors from 24 HER2+ early BC patients as a preliminary study. As with our sensitive and resistant models, we analyzed TGFBI methylation by bisulfite pyrosequencing in pre-treatment samples from patients who were complete responders and non-responders to trastuzumab-based therapy (Fig. 4a). With the non-responder patients, we also evaluated TGFBI methylation after treatment (post-treatment samples), i.e., as we did in our SKTR model. For the non-response patients, we obtained paired samples (pre-treatment and post-treatment samples) from 10 patients, pre-treatment samples only from 3 patients, and post-treatment samples only from 7 patients. For the patients with pathological complete response, only pre-treatment samples were evaluated. The limited number of samples obtained was, in part, due to the exhaustion of the sample during diagnostic procedures and the poor preservation of the DNA in the paraffin tissue (FFPE) blocks.

Our results showed similar pre-treatment TGFBI methylation levels (p = 0.651) in the patients with complete response to trastuzumab (6.45% ± 1.90) and in the non-responders (6.08% ± 1.51; Fig. 4b). These results indicate that the TGFBI methylation levels of the pre-treatment samples are not associated with the absence of response to trastuzumab. In contrast, the non-responsive patients showed significantly higher (p = 0.001) methylation levels of TGFBI in tumors following treatment with trastuzumab (30.26% ± 3.52) than before starting the therapy (6.08% ± 1.51), suggesting that acquired resistance to trastuzumab is associated with increased methylation levels of TGFBI. In particular, when considering the non-responsive patients with pre- and post-treatment paired samples, we observed a significant (p < 0.001) TGFBI promoter hypermethylation after trastuzumab (8 out of 10 = 80%) compared to the pre-treatment samples, indicating that the increase in TGFBI methylation levels is associated with trastuzumab resistance (Fig. 4c). Importantly, the ROC curve analysis with an AUC of 0.95 (p < 0.0001; 95% CI 0.803 to 0.996) allowed us to clearly differentiate the methylation levels of TGFBI between the pre- and post-treatment samples of the non-responsive patients. This result suggests that TGFBI could be a potential biomarker for monitoring trastuzumab response during HER2+ BC patients’ treatment (Fig. 4d). In addition, the Hosmer-Lemeshow test (chi-square = 89,606; p = 0.3456) suggests that the methylation of TGFBI has a certain ability to predict the resistance to trastuzumab when compared with an ideal model (Additional file 4: Figure S3). On the other hand, no significant association between TGFBI methylation levels before and after treatment and the clinical-histopathological characteristics was identified (Table 2).